Bispecific repurposed medicines targeting the viral and immunological arms of COVID-19

Effective agents to treat coronavirus infection are urgently required, not only to treat COVID-19, but to prepare for future outbreaks. Repurposed anti-virals such as remdesivir and human anti-inflammatories such as barcitinib have received emergency approval but their overall benefits remain unclear. Vaccines are the most promising prospect for COVID-19, but will need to be redeveloped for any future coronavirus outbreak. Protecting against future outbreaks requires the identification of targets that are conserved between coronavirus strains and amenable to drug discovery. Two such targets are the main protease (Mpro) and the papain-like protease (PLpro) which are essential for the coronavirus replication cycle. We describe the discovery of two non-antiviral therapeutic agents, the caspase-1 inhibitor SDZ 224015 and Tarloxotinib that target Mpro and PLpro, respectively. These were identified through extensive experimental screens of the drug repurposing ReFRAME library of 12,000 therapeutic agents. The caspase-1 inhibitor SDZ 224015, was found to be a potent irreversible inhibitor of Mpro (IC50 30 nM) while Tarloxotinib, a clinical stage epidermal growth factor receptor inhibitor, is a sub micromolar inhibitor of PLpro (IC50 300 nM, Ki 200 nM) and is the first reported PLpro inhibitor with drug-like properties. SDZ 224015 and Tarloxotinib have both undergone safety evaluation in humans and hence are candidates for COVID-19 clinical evaluation.

www.nature.com/scientificreports/ play a role suppressing viral proliferation in patients already infected at an early stage in disease and thus reduce overall disease burden 6 and have so far shown broad spectrum activity against several coronavirus variants 7,8 . Existing small molecules for treating COVID19 include the anti-viral RNA polymerase inhibitor remdesivir 9 (originally developed for Hepatitis C Virus) as well as compounds with anti-inflammatory and immunosuppressant effects such as dexamethasone 10 and baricitinib 11 . Unfortunately, none of these medicines have delivered meaningful benefits to patient populations, execpt dexamethasone which provides benefits in only the most advanced stages of the disease 10 , although combinations of remdesivir and baricitinib show promising results 11 . Thus there is strong motivation for the discovery of new therapeutics not only for the treatment of acute disease. Antivirals offer the potential of prophylaxis, reduction of transmissibility, treatment of unvaccinated patients and suppression of emergent coronaviruses. Considering the lengthy timescales required to develop and approve new therapeutic agents, repurposing of known drugs can potentially reduce the time to develop new treatments.
The SARS-CoV-2 genome encodes small molecule druggable targets including the main protease 12 (M pro ) encoded by non-structural protein 5 (nsp5) and the papain-like protease 8 (PL pro ) which is part of non-structural protein 3 (nsp3) 13 . As both proteins are essential for viral replication they present attractive targets for drug repurposing efforts. These cysteine proteases are encoded along with the other 14 nsps by the 5′-terminal open reading frame 1a/b (ORF1a/b), which takes up approximately two-thirds of the viral genome. This leads to the expression of the two large replicase polyproteins pp1a and pp1ab. M pro and PL pro are responsible for the proteolytic cleavage of pp1a and pp1ab which consist of nsps 1-11 and 1-16, respectively 14 . M pro cleaves at 11 sites releasing the functional nsps 4-16, while PL pro cleaves at 3 sites releasing nsps 1-3 15 . Additionally, PL pro acts as a deubiquitinase and deISGylase which may modulate the host anti-viral response via suppression of type-I interferon production 16 . Recent developments have seen a rationally designed M pro inhibitor enter clinical testing 17 .
To rapidly identify potential anti-coronavirus therapeutics, we undertook extensive experimental screens of the drug repurposing ReFRAME library, that consists of 12,000 therapeutic agents against M pro and PL pro to identify clinically-viable agents that may be repositioned to treat COVID-19 and future outbreaks. One compound, SDZ 224015, is a potent irreversible inhibitor of M pro , whilst a second, Tarloxotinib, is a sub micromolar inhibitor of PL pro .

Results
ReFRAME library screening and hit triage. The ReFRAME library comprises 12,000 molecules that have been previously approved for clinical investigation in humans, including all currently approved medicines 18 . Despite the overall quality and relevance of this library, it contains some older compounds with properties less attractive for modern drug discovery, such as polyphenol groups, flavonoids and catechols (sennosides) 19 , reactive Michael acceptors (oxantel) 20 , unattractive molecular weights and poor solubility, many of which can also cause assay interference 21 .
M pro and PL pro have the potential to exist in several conformationally distinct states, each of which may favour the binding of different inhibitors 22 . M pro substrate velocity titrations revealed evidence for catalytically distinct monomeric and dimeric forms 23 . Dimer dependent catalysis manifests as the observation of enzyme concentration dependent sigmoidal substrate velocity plots (Fig. 1a) as observed for M pro from SARS-CoV-1 24 . The parameters describing the midpoint or slope of the sigmodal substrate velocity plots displayed a bell-shaped relationship with enzyme concertation (Fig. 1b,c), whereas the maximum velocity displayed a sigmoidal relationship with enzyme concentration (Fig. 1d). These results strongly indicate dimerization may be induced either by increasing enzyme or substrate concentration, at high concentrations (> 300 nM) M pro spontaneously dimerizes and low concentration (< 3 nM) M pro behaves as a monomer. These data indicate the potential for inhibitors to bind to these catalytically distinct forms 25 as well as the potential for binding at the dimer interface 24 . Consequently, the M pro HTS assay was designed to balance these forms. PL pro was found to require the presence of high concentrations of anionic Hoffmeister salts to catalyse hydrolysis of small peptide substrates, but not large ubiquitin mimetics (Fig. 2). This indicated that the PL pro active site is not accessible in isotonic buffer, as previously observed for SARS-CoV-1 PL pro26 . This requirement may relate to the subcellular location of nsp3 expressed during viral infection. Nsp3 is expressed on the surface of the endoplasmic reticulum and in combination with nsp4 creates double membraned vesicles 27 . For the papain-like protease of the betacoronavirus murine hepatitis virus, PL pro did not process the viral polyprotein unless expressed on the ER membrane 28 . Thus, to maximise the discovery of inhibitors, the HTS was run in the presence of 0.8 M citrate.
In order to minimise the number of HTS false positives, confirmatory screens were run, followed by a counter screen of M pro and PL pro hits against one another, taking advantage that despite minimal homology both have nucleophilic cysteine containing active sites but recognise distinct peptide substrate sequences.
An initial single point screen of the ReFRAME library performed well for M pro and adequately for PL pro (Supplementary Fig. 1). Following single point screening, a hit triage consisting of repeat confirmation and selectivity counter screening ( Supplementary Fig. 2), twenty-one M pro and thirty-five PL pro compounds remained (Supplementary Table 1). After filtering to remove undesirable chemical structures, such as pan assay interference compounds etc., two novel M pro and a single PL pro hit were selected for further analysis.

SARS-CoV-2 M pro inhibitors.
The two selected M pro hits from the ReFRAME screen are compounds 1 and 4 (Fig. 3). 4 has an IC 50 of 30 nM (the biochemical limit of the assay), however contains a less attractive peptide backbone scaffold, whilst 1 exhibits a weaker IC 50 of 3 µM and structure consistent with drug-like absorption, distribution, metabolism, and excretion (ADME) properties 29 . 1 is a derivative of the investigational compound ABT-957 30 (2, Fig. 3, Supplementary Table 1) a calpain 1 & 2 inhibitor, differing only by the presence of a pyridinyl group instead of the ABT-957 cyclopropyl-group. This substitution is responsible for the M pro inhibition potency of 1 compared to 2 (IC 50 > 100 µM, Fig. 3). Inhibition of M pro does not seem to be a general property  Compound 4 is the investigational caspase 1 inhibitor prodrug SDZ-224015 31 , which is cleaved by esterases in vivo to yield 5, where the free aspartic acid is revealed (Fig. 3). In contrast to the exquisite potency of 4, 5 has limited potency for M pro , yielding only 50% inhibition at 100 µM. Inhibition of M pro did not seem to be a general property of caspase 1 inhibitors, as the tool tetrapeptide Ac-YVAD-AOM (6, Fig. 3) and the investigational caspase 1 drug belnecasan (7, Fig. 3) did not substantially inhibit M pro . 4 was confirmed to bind to M pro in an SPR assay, although due to the mechanism of action a K D cannot be reported ( Supplementary Fig. 3). Compound 4 is a suicide inhibitor which is cleaved by M pro , releasing a dichlorobenozic acid leaving group and forming an irreversible covalent adduct by reaction with the nucleophilic cysteine ( Supplementary Fig. 4).
Due to the presence of three esters, 4 is unstable in aqueous media so is unsuitable for the long incubations used in antiviral assays. However, when the dose was refreshed daily, 4 was found to be non-cytotoxic in HUH7 cells at concentrations at or below 10 µM and showed an antiviral effect of 50% at 10 µM (Fig. 4). Refreshing the dose more frequently is likely to further increase the apparent potency. SARS CoV-2 PL pro inhibitors. The ReFRAME screen revealed a single inhibitor, tarloxotinib as a potent PL pro inhibitor ( Table 1, 8) which has an IC 50 of 300 nM. 8 is also a prodrug, activated in hypoxic conditions in vivo by STEAP4 32 to yield the equipotent compound 9 (Table 1) which was designed to target the kinase domain of EGFR 33 . Whilst both 8 and 9 contain a 4-anilinoquinazoline core that is present in several approved drugs 34 , the tested related molecules proved to be less potent than tarloxotinib (Table 1, [10][11][12][13]. Analysis of the other 4-anilinoquinazoline approved medicines revealed the presence of the α,β-unsaturated amide warhead on 8 and 9 was not necessary or sufficient for activity, as 13 achieved weak activity without the warhead whilst the presence of a nitrile on the 3 position of 10 completely removes potency despite the presence of the warhead. These observations show that inhibition is not solely due to intrinsic reactivity of the molecules but requires specific molecular recognition.   www.nature.com/scientificreports/ 8 was found to be competitive with the peptide substrate ( Supplementary Fig. 6) and despite the conditions required for its discovery, the potency of 8 was not dependent on the use of high concentrations of anionic Hoffmeister salts ( Supplementary Fig. 7). In HUH7 cells 8 was non-cytotoxic at concentrations up to 10 µM, and showed an antiviral effect of 25% at 10 µM (Fig. 4).
Crystallography of the ReFRAME hits. X-ray crystal structures were attempted for compounds 1,4,5  Table 2). The M pro dimer is shown in ribbon representation with 1 and 5 bound at the active site ( Fig. 5a,b). Both 1 and 5 bind covalently to the catalytic cysteine (Cys145) with well-defined electron density and form hydrogen bonding networks with the M pro active site (Fig. 5c,d, Supplementary Fig. 8). Despite both 1 and 5 interacting with Cys145, they have substantially different binding modes. 5 extends from P1 to P5 of the M pro active site (Fig. 5b). By contrast 1 binds from P1 to P1′ across the region occupied by the catalytic cysteine, with the two inhibitor benzyl groups π-stacking together to fill the space of the P1′ pocket (Fig. 5b). Electron density for the 2,6-dichlorobenzoate leaving group of 5 was not observed, providing evidence of the proposed mechanism of inhibition (Supplementary Fig. 4 and 8). 1 forms electrostatic interactions with Gly143, Ser144, Cys145, and His41 as well as a water-mediated interaction with His164 whereas 5 makes electrostatic interactions with Gly143, Cys145, His163, His164, and Glu166 together with a water mediated interaction with Gln189 (Fig. 5). www.nature.com/scientificreports/ Active site plasticity is important in accommodation of the inhibitors. The P1′ pocket expands on binding of 1, with the alpha carbons of Thr25 and Thr26 shifting by 0.9 Å and 0.7 Å respectively. Similarly, the P5 pocket expands upon binding to 5, with Pro168 and Thr190 moving by 1.3 Å and 0.8 Å respectively. Additionally, there is a 1.9 Å shift between Ala191 in the 1 and 5 complexes. In both cases plasticity in response to ligand binding is also observed for the P2 pocket 35 (Fig. 6).
The pyridine of 1 and the aspartate of 5 extend into the P1 pocket and interact with His163 (Fig. 5c,d). Modelling suggests that the cyclopropyl ring sidechain of 2 is unable to make this interaction and as a consequence does not bind to M pro . Similarly, the acid of the aspartate in 5 is in close proximity to Glu166 residue which may cause a charge clash explaining the loss in potency of 5 compared to 4.
Prospects for molecular design. Combining the information from the M pro structures of 1 and 5 could be the starting point to design more potent, drug-like inhibitors. As inhibition of Caspase 1 would inhibit inflammation via suppression of the IL-1β, this could provide additional clinical benefit in the treatment of COVID-19 36,37 . Thus inhibitors which possess dual anti-inflammatory and antiviral properties may be desirable. A docking analysis 38,39 of the binding pose of 5 in M pro and the caspase-1 active site reveals multiple shared interactions, indicating that further, more drug-like molecules could be developed which share the potential dual anti-viral/ anti-inflammatory polypharmacology of SDZ 224015 ( Supplementary Fig. 9).
For PL pro , the 4-anilinoquinazoline core is one of the most common scaffolds for generation of tyrosine kinase inhibitors. Thus, it should be possible to rapidly expand from 8 and 9, to discover new, potentially more potent PL pro inhibitors with the potential to remove kinase activity all together, whilst retaining drug-like properties.

Discussion
The emergence of COVID-19 has emphasised the need for multiple approaches to tackle viral infections. To bridge the gap between the need to rapidly address a new disease and the time required to safely develop an entirely new medicine, repurposing existing drugs is an attractive alternative 40 .
There have been several clinical efforts to assess the benefits of existing drugs for COVID-19 treatment. Trials have broadly focused on repurposed anti-viral drugs to reduce infection such as remdesivir, as well as the use of existing anti-inflammatory and immunosuppressant compounds to help the body better manage its subsequent response to the infection. Virtual screening has also been used 23,35 , but has only identified boceprevir, an HCV protease inhibitor, as an M pro inhibitor 35 . This molecule was identified in our screen but has disappointing activity (IC 50 of 3 µM) (Supplementary Table 1).
By contrast, the results described here identify highly potent inhibitors of M pro and for the first time a potent inhibitor of PL pro with drug-like properties. This is also the first description of a non-antiviral molecule to show repurposed PL pro activity. Neither of the two M pro inhibitors discovered in this study were proposed by docking efforts. 4 is a suicide inhibitor which uses a complex mechanism that is difficult to predict 41 . Further, the conformation of 1 within the P1' pocket of M pro which is driven by intramolecular pi-pi stacking (Fig. 3c) is unusual and not readily predicted by in silico approaches 42 .
A recent crystallographic screen of 5000 compounds discovered several compounds which crystallised with M pro43 . One such hit was the EGFR inhibitor pelitinib but disappointingly it only subsequently shows micromolar activity in a cellular screen and was not determined to display significant biochemical inhibition of M pro in this study. In contrast to the M pro crystallographic screen which was also restricted to a single structural form, we were able to study both M pro and PL pro in solution where multiple conformations and oligiomeric forms are www.nature.com/scientificreports/ present. Pelitinib is compound 10 in our study, but our results show that the 4-aminoquinazoline class of EGFR inhibitors 8 and 9 are more promising; and importantly operate as potent PL pro , rather than M pro , inhibitors. By employing optimised screens, we specifically interrogated the two essential SARS-CoV-2 viral proteases, discovering compounds not identified in previous phenotypic screens despite possessing anti-viral activity. The ReFRAME collection has been screened in phenotypic viral-replication 44 assays. In spite of counter screens, without deconvolution, the results from phenotypic screens can artificially prioritise highly potent compounds such as transcription inhibitors and cytotoxic compounds that have undesirable mechanisms of action precluding therapeutic development 45 along with undervaluing the potency of viable compounds. Compounds that require revised assay protocols to observe activity, such as 4, were therefore not previously identified.
In vitro viral replication assays are of limited value for predicting the in vivo pharmacodynamics of candidate molecules 46 where multi-day assays often underestimate the true in vivo potency. For compounds such as 4, where a confounding factor is the aqueous stability of the molecule, in vitro data serve to support the mechanism rather than predict in vivo efficacy, where administration frequency, route and immune clearance would positively influence potency 47,48 .
Similarly, potencies of anti-viral activity can vary drastically depending on the methods used. A recent study of the PL pro tool inhibitor GRL-0617 8 saw potency vary by two orders of magnitude between biochemical (IC 50 of 2 µM), cytopathic-effect (30 µM), viral RNA detection (> 50 µM) and FFU (IC 50 > 100 µM) assays. Consequently, there is a prospect that the potencies of both 4 and 8 in vivo may be better than implied by the cytopathic-effect antiviral measure used in this work.
There have been three widespread outbreaks of fatal novel respiratory coronavirus mediated disease in the last two decades 49 . Retrospectively, the dangers of further outbreaks were evident following the first 50 . To avoid the debilitating effects of future coronavirus pandemics or even escape from immune protection, a range of treatments are necessary in which effective antiviral drugs will be a critical component. Protease inhibitors have been highly successful in combating other viral infections 51 . The high conservation of M pro and PL pro between the three strains of coronavirus which cause greatest impact on human health suggest that these are excellent target opportunities for developing small-molecule anti-viral therapeutics.
Anti-viral efforts aim to treat patients who are already infected and halt progression to severe disease 6 . This serves to reduce the burden of disease on fragile health care systems, but must also be employed alongside vaccination 3 and containment efforts 52 . Vaccination and containment serve to prevent the potential for infection, whereas anti-viral aim to treat those already infected. To be truly useful anti-viral medicines must be broad spectrum and stockpiled prior to an outbreak as suggested for influenza 53 .
Our studies describe the discovery of potent, drug-like inhibitors for both M pro and PL pro . These inhibitors display in vitro antiviral activity and have already been shown to be safe for clinical investigation for other therapeutic areas. Given their existing preclinical safety profiles these compounds have the potential for rapid progression towards a clinical setting.
The African monkey kidney cell line Vero E6-GFP was a gift kindly provided by M. van Loock, Janssen Pharmaceutica, Beerse, Belgium.
The hepatocellular carcinoma cell line Huh7 was a gift kindly provided by Ralf Bartenschlager, University of Heidelberg, Germany.
All compounds were obtained at a manufacturer specification of > 98% purity. Unless otherwise stated all other reagents were from Sigma Aldrich (Poole, UK).

Construct design and construction. The M pro coding sequence was codon optimised for expression in E.
coli and synthesised by Integrated DNA technologies (IDT). The M pro expression construct used for crystallization comprises an N-terminal GST region, an M pro autocleavage site, the M pro coding sequence, a hybrid cleavage site recognizable by 3C HRV protease and a C-terminal 6-Histidine tag 54 . The overall construct was flanked by In-Fusion compatible ends for insertion into BamHI-XhoI cleaved pGEX-6P-1 (Sigma). An additional M pro construct was generated with an extended 10-Histidine tag, for enhanced binding to the sensor surface in SPR assays. This construct was amplified by PCR from the above version, with the C-terminal primer incorporating a further 4-Histidines. The resulting amplicon was then inserted into BamHI-XhoI cleaved pGEX-6P-1 by In-Fusion cloning. The PL pro expression construct was similarly optimised and synthesised and comprised an N-terminal 10 Histidine tag followed by the PL pro sequence (Nsp3 region E746-K1060). This was then directly inserted into NcoI-HindIII digested pOPINF via In-Fusion compatible ends. pOPINF was a gift from Ray Owens (University of Oxford) 55  www.nature.com/scientificreports/ Protein expression of M pro with authentic termini. The plasmids were used to transform a competent E. coli expression cell line based on BL21(DE3)-R3-pRARE. The cells were plated on LB-agar plates containing 50 µg/ml carbenicillin and incubated overnight at 37 °C. The next day multiple colonies were picked and use to inoculate a series of consecutive starter cultures (LB, 50 µg/ml Carbenicillin) of 1 ml, 10 ml and 100 ml. At each stage the culture was grown to the exponential phase (OD 600 0.6-2, 200 rpm, 37 °C) before using the total volume of culture to inoculate the next, where the inoculate comprised 10% of the volume of the next culture in the series. Once 100 ml of exponential culture was achieved, 10 ml of this was used to inoculate 1 L of Auto Induction medium (Formedium, Terrific broth base including trace elements, prepared to manufacturer's instructions with addition of 10 ml glycerol and 50 µg/ml carbenicillin). Cultures were grown for 5 h at 37 °C, 200 rpm, followed by 15-20 h at 18 °C, 200 rpm. Cells were harvested by centrifugation and stored at -80 °C.
Protein purification of M pro with authentic termini for crystallographic analysis. Cells were resuspended in lysis buffer, 50 mM Tris pH 8, 300 mM NaCl, 10 mM imidazole, 0.03 μg/ml Benzonase, and lysed using an Emulsiflex homogeniser (3 passes, 30 kpsi, 4 °C). Insoluble material was removed by centrifugation (50,000 g, 4 °C). Tagged M pro protein was captured using Nickel-NTA (Takara His60 Superflow Resin) washed with 50 mM Tris pH 8, 300 mM NaCl, 25 mM imidazole, and eluted with 50 mM Tris pH 8, 300 mM NaCl, 500 mM imidazole. To remove the M pro poly-histidine tag, N-terminal His tagged HRV 3C protease was added to the eluted M pro fractions at a ratio of 1 mg 3C protease: 10 mg M pro . The mixture was dialysed overnight into 50 mM Tris pH 8, 300 mM NaCl, 0.5 mM TCEP at 4 °C and purified by reverse Nickel-NTA. Gel filtration was performed using a 16/600 Superdex S200 pg column (GE Healthcare) equilibrated in 50 mM Tris pH 8, 300 mM NaCl buffer. M pro was concentrated to 36 mg/ml using a centrifugal filter device with a 10 kDa molecular weight cut off prior to flash freezing using liquid nitrogen.
Expression and purification of M pro -His10. M pro -His10 was prepared as for M pro with authentic termini with the following modifications. The HRV 3C protease cleavage and reverse Ni-NTA steps were omitted. Instead, the Ni-NTA purified tagged M pro was dialysed into 50 mM Tris pH 8.5, 100 mM NaCl, 0.5 mM TCEP at 4 °C overnight. The dialysed sample was rapidly diluted with 50 mM Tris pH 8.5 buffer to achieve a final NaCl concentration of 25 mM. The protein was purified by anion exchange chromatography using a 5 ml HiTrap Q HP column (GE Healthcare) on a NaCl concentration gradient between 25 mM to 0.5 M NaCl. Ion exchange chromatography was followed by a gel filtration purification step as described above.
Expression and purification of cleaved PL pro . Cleaved PL pro was prepared as for M pro with authentic termini with the following modifications. A tunable T7 expression strain based on Lemo21 (DE3) was utilised in the expression of PLpro PL pro and 34 μg/ml chloramphenicol was added to all solid and liquid media to maintain its pLemo plasmid. 2 mM Rhamnose was included in the medium in the overnight and sub-culture stages. 0.5 mM Rhamnose was included in the auto induction medium. Instead of the HRV 3C protease, TEV protease was used to cleave the 10 Histidine tag of PL pro at the same mass ratio as described above. A 16/600 Superdex S75 pg column (GE Healthcare) was used for gel filtration chromatography.
Expression and purification of His10-PL pro . His10-PL pro was prepared as for cleaved PL pro with the following modifications. The TEV protease cleavage and reverse Ni-NTA steps were omitted. Instead, the Ni-NTA purified tagged PL pro was dialysed into 50 mM Tris pH 8.8, 100 mM NaCl, 0.5 mM TCEP at 4 °C overnight. The dialysed sample was rapidly diluted with 50 mM Tris pH 8.8 buffer to achieve a final NaCl concentration of 25 mM. The protein was purified by anion exchange using a 5 ml HiTrap Q HP column (GE Healthcare) with a NaCl concentration gradient between 25 mM to 0.5 M NaCl. The ion exchange chromatography was followed by a gel filtration purification step as described above. The plates were incubated for a further 60 min at room temperature; assays employed a Pherastar FSX plate reader (BMG Labtech, Aylesbury, UK) using a TAMRA filter set. Assay quality was established using Z' , where a low control consisted for substrate alone with balanced DMSO, and the high control was enzyme and substrate without inhibitors but balanced DMSO. The raw fluorescence data was converted to percent inhibition using the same controls as for Z' . www.nature.com/scientificreports/ M pro SPR assay. Cytiva Biacore S200 and T200 machines were used for all SPR experiments. Data were collected at a constant temperature of 20 °C. M pro -10His was captured on an NTA chip using standard protocols in running buffer: 20 mM Hepes (pH 7.5), 150 mM NaCl, 50 µM EDTA, 0.05% (v/v) Tween 20 and either 1 or 3% (v/v) DMSO at ~ 8500 RU. The compounds were screened at concentrations ranging from 23 nM to 50 µM adjusted appropriately for each compound, injecting from the lowest to highest concentrations. Scrubber 2 (Biologic software) was used to process and analyse SPR data. Kinetics were fitted using a 1:1 binding model with local Rmax for each concentration where required. Data for the inhibitors were referenced to those for a blank surface and blank injections to normalize for non-specific binding and drift. A DMSO calibration was run to remove excluded volume effect of binding responses between reference and target surface.
M pro protein observed mass spectroscopy. Protein MS-analyses were performed as described 56  Data collection and structure determination. All diffraction data were collected from crystals cryocooled to 100 K at Diamond Light Source. X-ray diffraction data for the M pro compound 1 complex were collected at beamline I04-1 at a wavelength of 0.9126 Å and data for the M pro compound 5 complex were collected at I24 at 0.9999 Å. Data were processed using Dials 57 via Xia2 58 . The datasets were phased using Molrep 59 and the M pro apo structure 60 . Ligand restraints were generated using GRADE (Global Phasing Ltd) and AceDRG 61 . Crystal structures were manually rebuilt in Coot 62 and refined using Refmac 63 and Buster 64 .
PL pro enzyme characterisation. 50  PL pro biochemical assay. Compound plates were prepared as for the M pro biochemical assay. 10 µL assay buffer (800 mM sodium citrate, 200 mM Tris, 0.5 mM TCEP, 0.05% (v/v) Triton X-100, pH 7.6) was added to the assay plates to solubilise the compound. 10 µL of PL pro at 75 nM in assay buffer, was added to the appropriate wells and incubated with the compounds for 60 min at room temperature. The reaction was started via the addition of 10 µL of the substrate [5-TAMRA]-VLRLRGG-[Lys(BHQ-2)]-amide at 6 µM in assay buffer. This resulted in final assay conditions of 25 nM PL pro and 2 µM substrate, with either 0.1% or 1% DMSO (v/v). The plates were incubated for a further 60 min at room temperature and the assay was read on a Pherastar FSX plate reader (BMG Labtech, Aylesbury, UK) using a TAMRA filter set. Assay quality was established as for the M pro biochemical assay. Data were converted to percent inhibition in the same manner as the M pro assay.
PL pro kinetic assays. Assays employed the same conditions as the PL pro inhibition assays, except the compound and substrate were prepared in the assay plate and the assay was started with an injection of 10 µL of enzyme using a Pherastar FSX plate reader. Fluorescence was measured every 15 s post injection using a TAMRA were plated with HUH7_mCherry cells at 6000 cells/well in 100 µl. The day after (Day 0), compound was added in a dilution series for concentration response studies. After two hours, addition of virus dilution (final MOI 0.004) was performed and plates were left for incubation at 37 °C, 5% CO2 for four days. Cytotoxicity was assessed in parallel using the same protocol, albeit without the addition of virus dilution. Plates were imaged on an Arrayscan XTI, Thermofisher.
Image acquisition and analysis. At day four post-infection, mCherry signal was captured using wide field fluorescence imaging by exciting at 560_25 nm and emitting with the BGRFRN filter set. A 5 X objective sufficed to capture 65-70% of an entire well on a 96well plate (4 pictures in total). The optimal exposure time was determined based on fluorescence intensity and was set on 0.09 s. A 2 × 2 binning was used and autofocus plane count was reduced to increase image acquisition speed. An image analysis protocol was developed in-house by using the SpotDetector bioapplication (Cellomics, Thermofisher). After background reduction on the raw image files, a fixed fluorescent intensity threshold was determined for the identification of mCherry cells. Afterwards, the number of fluorescent cells ('object count') was calculated per well and compared to the positive (cell control) and negative (virus) control.

Synthesis of SDZ-224015 (compound 4).
Commercially-sourced reagents (Sigma-Aldrich, Inc.; Fluorochem Ltd; Bachem AG) were used as received. Reactions were performed in anhydrous solvents (Sigma-Aldrich Inc.). Purifications, reaction work-ups, and extractions were performed using HPLC grade solvents (Sigma-Aldrich Inc.). A Stuart SMP-40 automated melting point apparatus was used to determine melting points (MP). A Bruker Tensor-27 Fourier transform infrared spectrometer was used for infrared (IR) spectroscopy. A Unipol (Schmidt Haensch) polarimeter was used for optical rotation (α) measurements. A Thermo Scientific Exactive mass spectrometer (ThermoFisher Scientific) operated in the positive ionization mode was employed for high-resolution mass spectrometry (HRMS) using electrospray ionization (ESI) mass spectrometry (MS); data are presented as a mass-to-charge ratio (m/z). A Bruker AVANCE AVIIIHD 600 spectrometer equipped with a 5 mm BB-F/ 1 H Prodigy N 2 cryoprobe was used for nuclear magnetic resonance (NMR) spectroscopy. Proton chemical shifts are reported in parts per million (ppm) downfield from tetramethylsilane, the residual protium in the NMR solvent is used as a reference (DMSO-d 6 : δ = 2.49 ppm). Carbon chemical shifts are reported in parts per million (ppm) in the scale relative to the NMR solvent (DMSO-d 6 : δ = 39.52 ppm). NMR data are reported as: chemical shift, multiplicity (m: multiplet, s: singlet, d: doublet, dd: doublet of doublets, t: triplet, q: quartet), coupling constant (J, Hz), and integration.

Research ethics statement.
All methods were carried out in accordance with relevant guidelines and regulations. All experimental protocols were approved by either, an internal research committee at Exscientia. Ltd, the ReFRAME committee or the CARE consortium. Viral swabs were obtained with prior patient's written informed consent for use in research. www.nature.com/scientificreports/